/gtfs-via-postgres

Process GTFS Static/Schedule by importing it into a PostgreSQL database.

Primary LanguageJavaScriptOtherNOASSERTION

gtfs-via-postgres

Import GTFS Static/Schedule datasets into a PostgreSQL database, to allow for efficient querying and analysis.

npm version binary build status Prosperity/Apache license minimum Node.js version support me via GitHub Sponsors chat with me on Twitter

  • ✅ handles daylight saving time correctly but retains reasonable lookup performance
  • ✅ supports frequencies.txt
  • ✨ joins stop_times.txt/frequencies.txt, calendar.txt/calendar_dates.txt, trips.txt, route.txt & stops.txt into views for straightforward data analysis (see below)
  • 🚀 is carefully optimised to let PostgreSQL's query planner do its magic, yielding quick lookups even with large datasets (see performance section)
  • ✅ validates and imports translations.txt
  • ✨ exposes (almost) all data via GraphQL using PostGraphile

Installation

npm install -g gtfs-via-postgres

Or use npx. ✨

There are also prebuilt binaries available.

Note: gtfs-via-postgres needs PostgreSQL >=14 to work, as it uses the WITH … AS NOT MATERIALIZED syntax. You can check your PostgreSQL server's version with psql -t -c 'SELECT version()'.

Getting Started

If you have a .zip GTFS feed, unzip it into individual files.

We're going to use the 2022-02-25 VBB feed as an example, which consists of individual files already.

wget --compression auto \
    -r --no-parent --no-directories -R .csv.gz \
    -P gtfs -N 'https://vbb-gtfs.jannisr.de/2022-02-25/'
# …
# Downloaded 14 files in 20s.
ls -lh gtfs
# 3.3K agency.csv
#  97K calendar.csv
# 1.1M calendar_dates.csv
# 2.5K datapackage.json
#  64B frequencies.csv
# 5.9K levels.csv
# 246B license
# 8.3M pathways.csv
#  49K routes.csv
# 146M shapes.csv
# 368M stop_times.csv
# 5.0M stops.csv
# 4.7M transfers.csv
#  16M trips.csv

Depending on your specific setup, configure access to the PostgreSQL database via PG* environment variables:

export PGUSER=postgres
export PGPASSWORD=password
env PGDATABASE=postgres psql -c 'create database vbb_2022_02_25'
export PGDATABASE=vbb_2022_02_25

Install gtfs-via-postgres and use it to import the GTFS data:

npm install -D gtfs-via-postgres
npm exec -- gtfs-to-sql --require-dependencies -- gtfs/*.csv | psql -b
# agency
# calendar
# CREATE EXTENSION
# BEGIN
# CREATE TABLE
# COPY 37
# …
# CREATE INDEX
# CREATE VIEW
# COMMIT

Importing will take 10s to 10m, depending on the size of the feed. On an M1 MacBook Air, importing the above feed takes about 4m; Importing the 260kb 2021-10-06 Amtrak feed takes 6s.

In addition to a table for each GTFS file, gtfs-via-postgres adds these views to help with real-world analysis:

  • service_days (materialized) "applies" calendar_dates to calendar to give you all days of operation for each "service" defined in calendar.
  • arrivals_departures "applies" stop_times/frequencies to trips and service_days to give you all arrivals/departures at each stop with their absolute dates & times. It also resolves each stop's parent station ID & name.
  • connections "applies" stop_times/frequencies to trips and service_days, just like arrivals_departures, but gives you departure (at stop A) & arrival (at stop B) pairs.
  • shapes_aggregates aggregates individual shape points in shapes into a PostGIS LineString.

As an example, we're going to use the arrivals_departures view to query all absolute departures at de:11000:900120003 (S Ostkreuz Bhf (Berlin)) between 2022-03-23T12:30+01 and 2022-03-23T12:35+01:

SELECT *
FROM arrivals_departures
WHERE station_id = 'de:11000:900120003'
AND t_departure >= '2022-03-23T12:30+01' AND t_departure <= '2022-03-23T12:35+01'
route_id route_short_name route_type trip_id date stop_sequence t_arrival t_departure stop_id stop_name station_id station_name
10148_109 S3 109 169035756 2022-03-23 00:00:00 19 2022-03-23 12:31:24+01 2022-03-23 12:32:12+01 de:11000:900120003:2:53 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)
10148_109 S3 109 169035899 2022-03-23 00:00:00 10 2022-03-23 12:33:06+01 2022-03-23 12:33:54+01 de:11000:900120003:3:55 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)
10162_109 S7 109 169128381 2022-03-23 00:00:00 19 2022-03-23 12:33:54+01 2022-03-23 12:34:42+01 de:11000:900120003:2:53 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)
10162_109 S7 109 169128495 2022-03-23 00:00:00 9 2022-03-23 12:30:36+01 2022-03-23 12:31:24+01 de:11000:900120003:3:55 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)
10223_109 S41 109 169054370 2022-03-23 00:00:00 21 2022-03-23 12:30:24+01 2022-03-23 12:31:12+01 de:11000:900120003:5:58 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)
10227_109 S42 109 169071882 2022-03-23 00:00:00 6 2022-03-23 12:30:30+01 2022-03-23 12:31:12+01 de:11000:900120003:5:59 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)
19040_100 RB14 100 178748721 2022-03-23 00:00:00 13 2022-03-23 12:30:00+01 2022-03-23 12:30:00+01 de:11000:900120003:1:50 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)
22664_2 FEX 2 178748125 2022-03-23 00:00:00 1 2022-03-23 12:32:00+01 2022-03-23 12:34:00+01 de:11000:900120003:4:57 S Ostkreuz Bhf (Berlin) de:11000:900120003 S Ostkreuz Bhf (Berlin)

translations

There are some …_translated views (e.g. stops_translated, arrivals_departures_translated) that

  • join their respective source table with translations, so that each (translatable) field is translated in every provided language,
  • add a …_lang column for each translated column (e.g. stop_name_lang for stop_name) that indicates the language of the translation.

Assuming a dataset with translations.csv, let's query all stops with a de-CE translation, falling back to the untranslated values:

SELECT
    stop_id,
    stop_name, stop_name_lang,
    stop_url,
FROM stops_translated
WHERE (stop_name_lang = 'de-CH' OR stop_name_lang IS NULL)
AND (stop_url_lang = 'de-CH' OR stop_url_lang IS NULL)

Usage

Usage:
    gtfs-to-sql [options] [--] <gtfs-file> ...
Options:
    --silent                  -s  Don't show files being converted.
    --require-dependencies    -d  Require files that the specified GTFS files depend
                                  on to be specified as well (e.g. stop_times.txt
                                  requires trips.txt). Default: false
    --ignore-unsupported      -u  Ignore unsupported files. Default: false
    --route-types-scheme          Set of route_type values to support.
                                    - basic: core route types in the GTFS spec
                                    - google-extended: Extended GTFS Route Types [1]
                                    - tpeg-pti: proposed TPEG-PTI-based route types [2]
                                    Default: google-extended
    --trips-without-shape-id      Don't require trips.txt items to have a shape_id.
    --routes-without-agency-id    Don't require routes.txt items to have an agency_id.
    --stops-without-level-id      Don't require stops.txt items to have a level_id.
                                  Default if levels.txt has not been provided.
    --stops-location-index        Create a spatial index on stops.stop_loc for efficient
                                    queries by geolocation.
    --schema                      The schema to use for the database. Default: public
    --postgraphile                Tweak generated SQL for PostGraphile usage.
                                    https://www.graphile.org/postgraphile/
Examples:
    gtfs-to-sql some-gtfs/*.txt | psql -b # import into PostgreSQL
    gtfs-to-sql -u -- some-gtfs/*.txt | gzip >gtfs.sql # generate a gzipped SQL dump

[1] https://developers.google.com/transit/gtfs/reference/extended-route-types
[2] https://groups.google.com/g/gtfs-changes/c/keT5rTPS7Y0/m/71uMz2l6ke0J

Some notable limitations mentioned in the PostgreSQL 14 documentation on date/time types:

For timestamp with time zone, the internally stored value is always in UTC (Universal Coordinated Time, traditionally known as Greenwich Mean Time, GMT). An input value that has an explicit time zone specified is converted to UTC using the appropriate offset for that time zone.

When a timestamp with time zone value is output, it is always converted from UTC to the current timezone zone, and displayed as local time in that zone. To see the time in another time zone, either change timezone or use the AT TIME ZONE construct […].

You can run queries with date+time values in any timezone (offset) and they will be processed correctly, but the output will always be in the database timezone (offset), unless you have explicitly used AT TIME ZONE.

With Docker

Instead of installing via npm, you can use the publictransport/gtfs-via-postgres Docker image:

# variant A: use Docker image just to convert GTFS to SQL
docker run --rm --volume /path/to/gtfs:/gtfs \
	publictransport/gtfs-via-postgres --require-dependencies -- stops.csv | psql -b

Keep in mind that psql -b will run outside of the Docker container, so your host machine needs access to PostgreSQL.

If you want to directly import the GTFS data from within the Docker container, you need add psql to the image and run it from inside. To do that, write a new Dockerfile that extends the publictransport/gtfs-via-postgres image:

FROM publictransport/gtfs-via-postgres
# add psql CLI tool
RUN apk add --no-cache postgresql-client
ENV PGPORT=5432 PGUSER=postgres PGPASSWORD=password
WORKDIR /gtfs
# pass all arguments into gtfs-via-postgres, pipe output into psql:
ENTRYPOINT ["/bin/sh", "-c", "env | grep PG; gtfs-via-postgres $0 $@ | psql -b"]
# start PostgreSQL DB in another container "db"
docker run docker run --name db -p 5432:5432 -e POSTGRES_PASSWORD=password postgis/postgis

# variant B: use Docker image to convert GTFS to SQL and import it directly
docker build -t import-gtfs . # build helper Docker image from Dockerfile
docker run --rm --volume /path/to/gtfs:/gtfs \
	--link db -e PGHOST=db \
	import-gtfs --require-dependencies -- stops.csv

Exporting data efficiently

If you want to export data from the database, use the COPY command; On an M1 MacBook Air, PostgreSQL 14 can export about 500k connections rows per second.

psql -c 'COPY (SELECT * FROM connections) TO STDOUT csv HEADER' >connections.csv

In the nested SELECT query, you can use features like WHERE, ORDER BY and LIMIT. Because psql passes on the exported data right away, you could stream it into another process.

Querying stops by location efficiently

If you want to find stops by (geo)location, run gtfs-via-postgres with --stops-location-index. This will create a spatial index on stops.stop_loc, so that most PostGIS functions & operators make use of it.

GraphQL support

The --postgraphile flag changes the SQL generated by gtfs-via-postgres slightly, so that you get a reasonably idiomatic GraphQL API out-of-the-box when running PostGraphile v4 on it:

# import data into PostgreSQL with PostGraphile tweaks
npm exec -- gtfs-to-sql -d --postgraphile -- gtfs/*.csv | psql -b

gtfs-via-postgres doesn't specify PostGraphile as a regular dependency, but as peerDependencies, in order to stay lightweight for users who don't need the GraphQL interface. Some versions of some package managers install unmet peer dependencies, some don't. Let's make sure that PostGraphile (and its plugins) are installed:

npm install \
    postgraphile@4.12 \
    @graphile-contrib/pg-simplify-inflector@^6.1 \
    @graphile/postgis@^0.2.0-0

The serve-gtfs-via-graphql helper script configures and runs PostGraphile. With NODE_ENV=development, it will

# listens on port 3000, this can be changed using $PORT
env NODE_ENV=development npm exec -- serve-gtfs-via-graphql

As an example for the GraphQL API, check out the test query or open the GraphiQL UI served at localhost:3000/graphiql.

Correctness vs. Speed regarding GTFS Time Values

When matching time values from stop_times against dates from calendar/calendar_dates, you have to take into account that GTFS Time values can be >24h and are not relative to the beginning of the day but relative to noon - 12h. (There are a few libraries that don't do this.)

This means that, in order to determine all absolute points in time where a particular trip departs at a particular stop, you cannot just loop over all "service dates" and add the time value (as in beginning_of_date + departure_time); Instead, for each date, you have to determine noon, subtract 12h and then apply the time, which might extend arbitrarily far into the following days.

Let's consider two examples:

  • A departure_time of 26:59:00 with a trip running on 2021-03-01: The time, applied to this specific date, "extends" into the following day, so it actually departs at 2021-03-02T02:59+01.
  • A departure time of 03:01:00 with a trip running on 2021-03-28: This is when the standard -> DST switch happens in the Europe/Berlin timezone. Because the dep. time refers to noon - 12h (not to midnight), it actually happens at 2021-03-28T03:01+02 which is not 3h1m after 2021-03-28T00:00+01.

gtfs-via-postgres always prioritizes correctness over speed. Because it follows the GTFS semantics, when filtering arrivals_departures by absolute departure date+time, it cannot filter service_days (which is calendar and calendar_dates combined), because even a date before the date of the desired departure date+time range might still end up within the range, when combined with a departure_time of e.g. 27:30:00; Instead, it has to consider all service_days and apply the departure_time to all of them to check if they're within the range.

However, values >48h are really rare. If you know (or want to assume) that your feed does not have arrival_time/departure_time values larger than a certain amount, you can filter on date when querying arrivals_departures; This allows PostgreSQL to reduce the number of joins and calendar calculations by a lot.

For example, when querying all absolute departures at de:11000:900120003 (S Ostkreuz Bhf (Berlin)) between 2022-03-23T12:30+01 and 2022-03-23T12:35+01 within the 2022-02-25 VBB feed, filtering by date speeds it up nicely (Apple M1, PostgreSQL 14.2):

station_id filter date filter query time nr of results
de:11000:900120003 none 230ms ~574k
de:11000:900120003 2022-03-13 >= date < 2022-04-08 105ms ~51k
de:11000:900120003 2022-03-23 >= date < 2022-03-24 55ms ~2k
de:11000:900120003 2022-03-22 > date < 2022-03-24 55ms ~2k
none none 192s 370m
none 2022-03-13 >= date < 2022-04-08 34s ~35m
none 2022-03-22 > date < 2022-03-24 2.4s ~1523k

Performance

With all use cases I could think of, gtfs-via-postgres is reasonably fast. If there's a particular kind of query that you think should be faster, please open an Issue!

The following benchmarks were run with the 2022-07-01 VBB GTFS dataset (41k stops, 6m stop_times, 207m arrivals/departures) on an M1 laptop; All measurements are in milliseconds.

query avg min p25 p50 p75 p95 p99 max iterations 
SELECT *
FROM stops
ORDER BY ST_Distance(stop_loc::geometry, ST_SetSRID(ST_MakePoint(9.7, 50.547), 4326)) ASC
LIMIT 100
 16.84 16.421 16.74 16.77 16.81 17.25 20.03 20.631 100 
SELECT *
FROM arrivals_departures
WHERE route_short_name = 'S1'
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
 27.77 26.994 27.61 27.74 27.85 28.47 29.86 30.14 100 
SELECT *
FROM arrivals_departures
WHERE station_id = 'de:11000:900100001' -- S+U Friedrichstr. (Berlin)
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
 275.65 272.239 273.74 275.29 276.59 280.18 281.45 281.923 40 
SELECT *
FROM arrivals_departures
WHERE station_id = 'de:11000:900100001' -- S+U Friedrichstr. (Berlin)
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
AND stop_sequence = 0
 202.48 199.638 200.39 201.55 204 207.88 211.66 212.071 50 
SELECT *
FROM arrivals_departures
WHERE stop_id = 'de:11000:900100001::4' -- S+U Friedrichstr. (Berlin)
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
 6.2 6.076 6.1 6.12 6.15 6.79 7.63 8.456 100 
SELECT *
FROM arrivals_departures
WHERE trip_id = '168977951'
AND date > '2022-08-08' AND date <= '2022-08-09'
 2.14 2.041 2.06 2.14 2.15 2.26 2.53 4.114 100 
SELECT count(*)
FROM arrivals_departures
WHERE stop_id = 'de:11000:900100001::4' -- S+U Friedrichstr. (Berlin)
 61.79 60.952 61.26 61.51 61.85 64.31 64.66 65.298 100 
SELECT count(*)
FROM arrivals_departures
WHERE stop_id = 'definitely-non-existent'
 2.08 2.047 2.06 2.06 2.07 2.11 2.31 2.451 100 
SELECT *
FROM arrivals_departures
WHERE t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
 1577.91 1555.598 1561.93 1572.38 1592.68 1604.32 1606.46 1606.992 10 
SELECT *
FROM connections
WHERE route_short_name = 'S1'
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
 85.27 78.838 80.98 82.99 91.25 93.18 94.78 94.899 100 
SELECT *
FROM connections
WHERE from_station_id = 'de:11000:900100001' -- S+U Friedrichstr. (Berlin)
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
 306.27 297.505 300.15 302.1 311.19 319.82 333.09 335.374 40 
SELECT *
FROM connections
WHERE from_station_id = 'de:11000:900100001' -- S+U Friedrichstr. (Berlin)
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
AND from_stop_sequence = 0
 231.07 224.655 229.2 230.86 231.86 239.84 247.15 253.035 50 
SELECT *
FROM connections
WHERE from_stop_id = 'de:11000:900100001::4' -- S+U Friedrichstr. (Berlin)
AND t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
 12.6 12.084 12.35 12.46 12.67 13.45 15.35 15.992 100 
SELECT *
FROM connections
WHERE trip_id = '168977951'
AND date > '2022-08-08' AND date <= '2022-08-09'
 3.17 3.057 3.09 3.12 3.15 3.46 4.18 4.709 100 
SELECT count(*)
FROM connections
WHERE from_stop_id = 'de:11000:900100001::4' -- S+U Friedrichstr. (Berlin)
 92.47 91.159 91.56 92.17 92.78 94.72 96.47 96.574 100 
SELECT count(*)
FROM connections
WHERE from_stop_id = 'definitely-non-existent'
 18.35 17.877 18.02 18.12 18.44 19.31 20.66 21.05 100 
SELECT *
FROM connections
WHERE t_departure >= '2022-08-09T07:10+02' AND t_departure <= '2022-08-09T07:30+02'
AND date > '2022-08-08' AND date <= '2022-08-09'
ORDER BY t_departure
LIMIT 100
 9050.33 9001.322 9013.13 9033.42 9078.03 9119.54 9132.11 9135.253 7

Related Projects

There are some projects that are very similar to gtfs-via-postgres:

Node-GTFS

Node-GTFS (gtfs npm package) is widely used. It covers three use cases: importing GTFS into an SQLite DB, exporting GTFS/GeoJSON from it, and generating HTML or charts for humans.

I don't use it though because

  • it doesn't handle GTFS Time values correctly (1/2, checked on 2022-03-01)
  • it doesn't always work in a streaming/iterative way (1/2, checked on 2022-03-01)
  • sometimes does synchronous fs calls (1/2, checked on 2022-03-01)

gtfs-sequelize

gtfs-squelize uses sequelize.js to import a GTFS feed and query the DB.

I don't use it because

  • it doesn't handle GTFS Time values correctly (1/2, cheked on 2022-03-01)
  • it doesn't provide much tooling for analyzing all arrivals/departures (checked on 2022-03-01)
  • some of its operations are quite slow, because they fetch relatved records of a record via JS instead of using JOINs

gtfs-sql-importer

There are several forks of the original outdated project; fitnr's fork seems to be the most recent one.

The project has a slightly different goal than gtfs-via-postgres: While gtfs-sql-importer is designed to import multiple versions of a GTFS dataset in an idempotent fashion, gtfs-via-postgres assumes that one (version of a) GTFS dataset is imported into one DB exactly once.

gtfs-via-postgres aims to provide more tools – e.g. the arrivals_departures & connections views – to help with the analysis of a GTFS dataset, whereas gtfs-sql-importer just imports the data.

other related projects

  • gtfsdb – Python library for converting GTFS files into a relational database.
  • gtfspy – Public transport network analysis using Python and SQLite.
  • GTFS Kit – A Python 3.6+ tool kit for analyzing General Transit Feed Specification (GTFS) data.
  • GtfsToSql – Parses a GTFS feed into an SQL database (Java)
  • gtfs-to-sqlite – A tool for generating an SQLite database from a GTFS feed. (Java)
  • gtfs-lib – Java library & CLI for importing GTFS files into a PostgreSQL database.
  • gtfs-schema – PostgreSQL schemas for GTFS feeds. (plain SQL)
  • markusvalo/HSLtraffic – Scripts to create a PostgreSQL database for HSL GTFS-data. (plain SQL)

License

This project is dual-licensed: My (@derhuerst) contributions are licensed under the Prosperity Public License, contributions of other people are licensed as Apache 2.0.

This license allows you to use and share this software for noncommercial purposes for free and to try this software for commercial purposes for thirty days.

Personal use for research, experiment, and testing for the benefit of public knowledge, personal study, private entertainment, hobby projects, amateur pursuits, or religious observance, without any anticipated commercial application, doesn’t count as use for a commercial purpose.

Get in touch with me to buy a commercial license or read more about why I sell private licenses for my projects.

Contributing

If you have a question or need support using gtfs-via-postgres, please double-check your code and setup first. If you think you have found a bug or want to propose a feature, use the issues page.

By contributing, you agree to release your modifications under the Apache 2.0 license.